Published June 28, 2026 | Version v1

Dataset and Error Analysis for the Evaluation of AI-Mediated Criminal Law Translation (DE/EN to EL)

  • 1. ROR icon Ionian University

Description

Abstract

This dataset contains the empirical data and error analysis generated for the research study titled "When AI Mistranslates Justice: Ethical, Legal, and Professional Risks of AI-Mediated Criminal Law Translation". It evaluates the performance of machine translation engines (DeepL, eTranslation) and Large Language Models (GPT-5.5, Claude) when translating sensitive institutional and criminal law terminology between German/English and Greek.

The dataset investigates systemic incongruence, stylistic flattening, register shifts, and terminological displacement (conceptual substitution with common law equivalents), highlighting the algorithmic legal hegemony in translated criminal proceedings.

File Structure & Contents
The dataset consists of the following comma-separated (CSV) files exported from the primary research spreadsheet:

1. Corpus & LLM Outputs (Dimensions: Columns A to BB, Rows 1 to 111)

  • Contains the term identification, thematic categories, reference terminology (Common Law / Civil Law systems), and the direct raw outputs generated by DeepL, eTranslation, GPT-5.5, and Claude.
  • Includes qualitative error tagging for English mediation, stylistic flattening, register shifts, and terminological displacement. 

2. Coding Guide

  • The methodological codebook defining the error categories, tags, and linguistic/legal variables applied during the manual and automated analysis of the outputs.

3. Summary Statistics

  • Aggregated quantitative metrics and rates regarding English mediation, stylistic flattening, and terminological displacement across different legal domains (Criminal Law, Civil Law, Procedural Law, Public & Constitutional Law).

Methodology & Variables

  • Language Pairs: German to Greek (DE → EL) and English to Greek (EN → EL).
  • Analyzed Terms: 105 legal terms per engine/model.
  • Evaluation Framework: Critical Discourse Analysis (CDA) and legal translation theory (anisomorphism and institutional asymmetry).


License & Citation
This dataset is made available under the Creative Commons Attribution 4.0 International (CC-BY 4.0) license. Please cite the associated publication/monograph or this Zenodo entry when utilizing these data for secondary analysis or comparative studies.

 

Files

Additional details

Dates

Submitted
2026-06-28